ModelMath-Shepherd PRM

Math-Shepherd

An open-source process reward model (PRM) designed for mathematical reasoning tasks. It is trained using automatically constructed process-level annotations and assigns quality scores to each intermediate step of a reasoning chain.

Timeline (last 90 days)

Oct 2

Experiments were conducted on Qwen2.5-7B-Instruct using Math-Shepherd as the PRM, tested on the MATH500 dataset (n=500, three random seeds).

Unverified50%

All Facts (1)

Unverified

Experiments were conducted on Qwen2.5-7B-Instruct using Math-Shepherd as the PRM, tested on the MATH500 dataset (n=500, three random seeds).

50%

Source Articles

Why Does PRM-Pruned Fragment Grafting Fail? An Empirical Analysis Across Three Reasoning ModelsOct 2